I'd disagree with the premise here. Sticking a blue filter on the front of your lens to balance tungsten light to a daylight sensor isn't going to do anything to improve the blue channel noise. In fact all it will do is reduce the amount of light falling on the red and green sensors, thus lowering the signal-to-noise ratio in the red and green channels.
Digital 'amplification' doesn't introduce noise, either. It will amplify existing noise as well as amplifying signal, leaving the signal-to-noise ratio unchanged. Applying analog amplification on the sensor before the A-to-D stage, however, will introduce some additional noise to the signal, thus decreasing the signal- to-noise ratio. I don't believe any sensor is capable of applying different analog gains to the separate colour channels (switching transients would be a nightmare at typical pixel scan-out rates); analog gain can be used to change the overall sensitivity of the sensor (although only to a limited degree), uniformly in R/G/B. The only artifact that analog pre-amplification helps with is in the reduction of quantization effects. These are only noticeable in the darkest parts of the image. With a 12-bit sensor and a properly exposed image you're going to get more than ten bits of signal in the brightest parts of the image, even with a multiplier applied (a typical white balance multiplier will be in the range of 0.85 - 2.2, according to my observations). That's enough for quite a bit of post-exposure manipulation before any effects show up in an image after conversion to 8 bits per component (which is all that most displays and printers are capable of accepting). > Hi, > I have few thoughts on white balance with DSLRs, and I would like > to discuss them to hear your opinions. > > I have two sort-of questions: > > 1) White balancing is IMO done by increasing the gain of R or B > channel. At least DCRAW and most OEM raw conversion > software works in this way. I don't know, however, if in-camera > jpeg or raw WB is done after the AD conversion, or somehow before > it. If it was done before (on the _analog_ signal), setting > correct WB in camera would be better than just correcting it in raw > conv. software in computer. If it was done on the already AD > converted signal (_digital_, quantitised signal), there would be no > difference between in-camera WB and raw conv. software WB correction. > (of course, in-camera WB is still useful when shooting JPG). > > In the first case, correctly setting WB in camera would probably > leave the RAW file less colour noisy than with correcting wrong WB > in raw conv. software. > > Anybody knows more? Or the answer? Or a way to test it? This is > pretty important if one wants the cleanest possible files. > > 2) IMO, every chip & its RGB filter set is calibrated for one type of light, > probably D5000 or D6500 daylight. If the scene is lit by another > light source, the camera (software) must increase the gain of some > channels to get the balance right. This means, when shooting under > tungsten light with ~3200K, setting the WB (either in camera or in > software, see [1]) for 3200K will mean it (camera or software) is > increasing the gain of the BLUE channel to make it look like ~5000K > lit scene. As all the channels have fixed quantum efficiency, > increasing gain means increasing noise as well. Increasing noise in > the blue channel, in our example. Thus, an indentical scene lit by > 5000K light in one case, and lit by 3200K light but balanced by the > DSLR to 5000K light in the other case... the second case will have > quite more blue channel noise! Right? > > The point I want to make, WHITE BALANCE is not miraculous. For best > results, we should use light temperature conversion filters on the > lens or on the light to bring the light temp back to the one the > camera is calibrated for (probably, daylight ~5000K). Otherwise, we > get lesser image quality in way of increased colour noise in some > channels. > > It works the same with colour negative film. You can correct it > during printing, but at the expense of some quality. > > So far, my real world experience seems to support these ideas. The > shots I have under tungsten light have more blue channel noise, > than the ones under daylight. > > Of course, we can decrease colour noise digitally, but even with > "smart" software, the result is the well-known sport-photography > bleached colour details look. Small colour details like subjects' > eye colour, pink lips, etc... all lose more or less saturation as > the "smart" filtering software mistakes them for noise. > > If anybody disagrees, please write me. I would be glad to be > proven wrong on this issue. > > Good light, > Frantisek Vlcek >
